Integrating indigenous local knowledge and species distribution modeling to detect wildlife in Somaliland

Abstract

Abstract. With wildlife populations in decline, understanding their distributions across the landscape are needed for management and conservation efforts, particularly in remote or hazardous regions. We used indigenous local knowledge to inform species distribution models (SDMs) to predict the distribution of 38 wildlife species historically documented in Somaliland, one of the most isolated, data-poor regions in Africa. We conducted 195 interviews with agro-pastoral men and women in 2016 and 2017 throughout Somaliland based on the environmental range captured by performing a Multivariate Environmental Similarity Surface (MESS) analysis. Respondents were asked if each species (1) occurred in the vicinity, and if they did, (2) whether it was common, rare, extirpated, or unknown. We tested two commonly used SDMs, Maxent and boosted regression trees (BRTs), to map the potential distribution of wildlife using interview data. We present detailed case studies of two species of high conservation priority: cheetah (Acinonyc jubatus) and African wild ass (Equus africanus somaliensis). More than half of the respondents reported the presence of 25 of the wildlife species in the survey. Additionally, we developed Maxent models for all surveyed wildlife known to develop a species richness map. The least reported species known to have inhabited Somaliland in recent decades were lion (Panthera leo), African wild ass, and Somali wild dog (Lycaon pictus somalicus). Surprisingly, 177 of respondents reported the presence of cheetah, which was presumed by many to be extirpated from Somaliland. Given the high number of reports, coupled with the BRT model results and evaluations, it may be that cheetahs persist throughout Somaliland. Unfortunately, our results also suggest that African wild ass may have been extirpated from Somaliland. Our models carry the uncertainties of the raw data, but provide habitat suitability maps to guide future targeted surveys. Our methods and results demonstrate that integrating indigenous local knowledge with SDMs may be an essential tool for the conservation of wildlife in data-poor regions.

Connecting Top-Down and Bottom-Up Approaches in Environmental Observing

“…A solution in such regions may be to connect knowledge of local resource users with remote sensing of vegetation and species distribution models. In 2016 and 2017, Evangelista and colleagues ( 2018 ) conducted 195 interviews with agropastoral men and women in Somaliland near the Horn of Africa and collected presence and absence information for 38 species of wildlife through interviews with goat herders who were shown photographs of relevant species. Remote sensing data on environmental variables and the information on where species have been recorded were used to draw maps showing the distribution of similar environments (based on 12 environmental predictor variables), thereby predicting the potential distribution of each species.…”

“…Both maximum entropy and boosted regression tree models in conjunction with the interviews show that African wild asses are confined to about a third of the country. Additional information from the interviews suggests that the population is dangerously low or extirpated from Somaliland (Evangelista et al 2018 ). Photograph: Mark D. Phillips, Science Photo Library.…”

Including indigenous knowledge in species distribution modeling for increased ecological insights

“…2016; Evangelista et al. 2018), local knowledge of species distribution patterns (Zhang & Vincent 2017), and application of expert understanding of species range boundaries to constrain the predictions of a SDMs that are parameterized with point occurrence records (Merow et al. 2017).…”

“…There will be no one optimal SDM technique for all IK models; rather, the best technique will depend on the research question and application (Elith & Graham 2009). There may be opportunities to incorporate IK in SDM methods that use local spatial knowledge, such as guidance in the collection of georeferenced presences points for wildlife (Luizza et al 2016;Evangelista et al 2018), local knowledge of species distribution patterns (Zhang & Vincent 2017), and application of expert understanding of species range boundaries to constrain the predictions of a SDMs that are parameterized with point occurrence records (Merow et al 2017). There are also opportunities for nonspatial IK to contribute to ecological modeling that incorporates expert knowledge (e.g.…”

Threat‐Based State and Transition Models Predict Sage‐Grouse Occurrence while Promoting Landscape Conservation

“…We used the Software Assisted Habitat Modeling (SAHM) (Morisette et al 2013) program developed by the U.S. Geological Survey to generate fit statistics between models developed with traditional landcover variables (Table 3) and those built using the TBSTM to represent biotic habitat components (Table 1). We specifically chose the SAHM modeling interface to R (R Core Team 2018), because the SAHM framework has been widely tested, is peer reviewed (Morisette et al 2013), and research that used SAHM has been extensively published in a wide variety of journals (Luo et al 2015, Evangelista et al 2018, Jarnevich et al 2018. We used the standard model fitting procedure in SAHM for generalized linear models (Young 2012).…”